set_backend

aimet_torch.quantization.set_backend(name)

Set global backend for quantization operations. Choices: [“triton”, “torch_builtins”]

Return type:

_ContextManager

Example

>>> # Temporarily set backend to triton
>>> with aimet_torch.quantization.set_backend("triton"):
...     aimet_torch.quantization.affine.quantize(
...         torch.arange(0, 1, step=0.1), torch.tensor(0.005), torch.tensor(0), 0, 255,
...     )
...
tensor([  0.,  20.,  40.,  60.,  80., 100., 120., 140., 160., 180.])
>>> # Permanently set backend to triton
>>> aimet_torch.quantization.set_backend("triton")
>>> aimet_torch.quantization.affine.quantize(
...     torch.arange(0, 1, step=0.1), torch.tensor(0.005), torch.tensor(0), 0, 255,
... )
...
tensor([  0.,  20.,  40.,  60.,  80., 100., 120., 140., 160., 180.])